The Data Must Flow: How Model Context Protocol (MCP) Accelerates AI Development [Webinar]

IT Training and Development Resources
Webinars
The Data Must Flow: How Model Context Protocol (MCP) Accelerates AI Development [Webinar]

Learn how to leverage Model Context Protocols (MCP) to overcome AI's data access limitations, standardize interactions, and accelerate AI development with real-world examples and a comparison to other frameworks. You will also gain insights into MCP's core benefits, adoption outlook, and future implications for AI systems.

How Model Context Protocols (MCP) Accelerates AI Development

Recorded June 23, 2025

While language models generate impressive results, they lack access to the live tools and real-world data sources that teams actually rely on to get work done. Every new integration (including connecting an AI to your CRM, knowledge base, or internal API—requires custom wiring) creates bottlenecks and inconsistency across teams.

The Model Context Protocol (MCP), introduced by Anthropic, proposes a solution: a standardized, vendor-neutral way for AI systems to discover and invoke tools and resources. Think of it as a kind of USB-C for AI—a consistent plug that allows any MCP-compatible model to interface with any MCP-exposed tool or data source. While still early in adoption, MCP offers the potential to unlock flexible, secure, and reusable integrations, reducing developer overhead and empowering more capable, enterprise-ready AI assistants.

What You'll Learn:

The problem MCP addresses: Why most AI systems today can't reach the data and tools that matter.
What MCP is: A high-level overview of the protocol, it's architecture, and how it works.
Core benefits: Standardization, discoverability, security, and future-proofing.
How MCP compares to frameworks like LangChain and OpenAI's Agents SDK.
How an AI agent can use MCP to access a weather forecast tool (real-world demo).
Adoption outlook: Who's using MCP now, what's still missing, and what to watch for.

WEBVTT

1
00:00:02.900 --> 00:00:19.950
Alex M.: Great. All right. I'm seeing some people trickle in. So while everyone's sort of joining today's session, welcome! Hello! My name is Alexandra, like I said, I'm going to be sort of the Mc for today's session. And the session being the data must flow. How model context protocol accelerates

2
00:00:19.950 --> 00:00:40.219
Alex M.: AI development. This session today is actually going to be centered around sort of the trending conversations you may have been seeing regarding model context protocol or Mcp, and how it could potentially be beneficial and adopted within AI systems. So before we begin today's sessions, I just kind of want to go over some housekeeping rules first, st

3
00:00:40.220 --> 00:01:04.699
Alex M.: as you might have noticed, mics and audio are going to be disabled for all attendees. However, it's, you know, completely meant to be an informative session, so you can direct any questions to the chat, I will be facilitating it. As Kevin is presenting today, we may not be able to read all of the current questions. So just please bear with us and be patient. We'll ensure we save some time at the end for any of the pending questions.

4
00:01:04.709 --> 00:01:25.999
Alex M.: and lastly, I will be sending a follow up email to everyone here today. It's going to include today's webinar recording. It's going to include a lab document as well and a quick post Webinar survey, where you have a chance to win some discounts on training upon submission. So please make sure to check that out. That will be sent out tomorrow.

5
00:01:26.080 --> 00:01:55.389
Alex M.: Now with that, said I. Just, wanna, you know, briefly introduce ascendant learning as a company very quickly for those who may not be familiar. So ascendant learning actually provides it. Training. We cover a wide variety of organizational needs at essentially every skill level. You know, we offer vendor authorized training and certifications for individuals. We also offer customized courses for teams, and we can develop that to, you know, accomplish project end goals.

6
00:01:55.390 --> 00:02:03.870
Alex M.: And then we also even specialize in, you know, generating tailored programs to organizations at scale.

7
00:02:03.930 --> 00:02:19.840
Alex M.: In addition to all of that that we do, we do have a large library of complementary resources. You know, much like today's webinar. Actually, we do tend to see requests asking us to run a session like today, internally for organizations.

8
00:02:19.940 --> 00:02:43.260
Alex M.: webinars like today's. These are great ways for organizations to learn about AI cloud or a myriad of like other it topics with business goals in mind. So that's that's something you might be interested in. I'm gonna throw a contact email in the chat for everyone and we can get in touch, you know, after this call with one of our instructors to to discuss that further.

9
00:02:43.260 --> 00:02:51.789
Alex M.: but with that said, I don't want to take up too much of the time for today's session. I'm actually going to stop sharing, and I'm going to pass it over to our host for today. Kevin Martin.

10
00:02:53.850 --> 00:02:59.100
Kevin Martin: Hey? I'm Kevin nice to meet you all. I'm gonna be running

11
00:02:59.270 --> 00:03:02.960
Kevin Martin: the session today. Hold on! Let me get my screen ready.

12
00:03:03.110 --> 00:03:05.649
Kevin Martin: I'm gonna be run running our session today.

13
00:03:08.620 --> 00:03:12.160
Kevin Martin: There we go share.

14
00:03:12.450 --> 00:03:18.449
Kevin Martin: I'll just share my whole desktop over here because we're gonna end up going back and forth

15
00:03:18.570 --> 00:03:23.440
Kevin Martin: between a few tools. But

16
00:03:24.090 --> 00:03:29.526
Kevin Martin: before we get into it just very brief. Bio

17
00:03:30.220 --> 00:03:47.440
Kevin Martin: I am. I'm Kevin Martin. I've been working in the Ml. And AI. Space for 5 ish years. Now, I have a master's in data science from Uc. Berkeley, and I'm still

18
00:03:47.590 --> 00:04:11.510
Kevin Martin: involved in some teaching there. I started off as as a ta in the statistics program. And then I got involved in the machine learning program. And I'm like, still helping run the statistics class there in addition to you know, developing courses with with Ascendian. And

19
00:04:11.860 --> 00:04:28.169
Kevin Martin: you know, teaching this webinar to y'all. So like like, Alex said, we're going to talk about model context protocol today, specifically and kind of how that fits into our overall picture of large language models and generative. AI,

20
00:04:30.120 --> 00:04:38.450
Kevin Martin: so yeah, overall like high level stuff. We're going to talk about tools in general, in

21
00:04:38.960 --> 00:04:48.119
Kevin Martin: our our generative AI space and what they are and why we need them. The issues they address. We'll kind of get into

22
00:04:49.360 --> 00:04:58.530
Kevin Martin: the issues with our 1st generation set of tools for generative AI, and and

23
00:04:58.730 --> 00:05:07.180
Kevin Martin: what limits they had, and how model context protocols is, is really

24
00:05:09.490 --> 00:05:24.310
Kevin Martin: fixing those the shortcomings. We'll talk about like, okay, actually, what is this model context? Protocol? We'll we'll talk about where it fits in. And the adoption outlook for it because we're all interested in that. Right?

25
00:05:27.754 --> 00:05:44.549
Kevin Martin: Okay, so what are Llm tools? Why do we need them? So our our basic large language models are, chat, gpt, Claude Gemini. They're they're useful, right? We can.

26
00:05:44.970 --> 00:05:58.579
Kevin Martin: They're useful on their own. We can go and ask them like, Hey, how do I bake a cake? And it will give us a pretty good idea of. You know how to bake a cake, or how to build a house, or

27
00:05:59.330 --> 00:06:08.030
Kevin Martin: Hey, how would I send a nicely worded email to my boss asking for a raise. It will. It will do all of these

28
00:06:08.370 --> 00:06:23.540
Kevin Martin: things. It knows it, it it knows how to do them right. But it doesn't actually have the ability to go out and interact and do the things that we're asking for. Right? So if we ask

29
00:06:23.650 --> 00:06:33.640
Kevin Martin: maybe more germane to this class. The data must flow right. Hey? I want to. I have a company database I want to pull out

30
00:06:33.890 --> 00:06:53.639
Kevin Martin: like what were our total sales last quarter. It can tell you you can go and kind of work as a middleman between the database and the model. Right? So you can go. Okay, we have these tables. Each table has these columns and these types, and you can kind of go back and forth to get it done. But

31
00:06:54.264 --> 00:07:08.995
Kevin Martin: the model doesn't actually have the ability to reach out and do the thing right. It's kind of like this octopus stuck in its tank, and we want to give it these little arms so that it can go and do the thing. It can go and query our database?

32
00:07:09.780 --> 00:07:11.046
Kevin Martin: So yeah.

33
00:07:11.980 --> 00:07:21.596
Kevin Martin: By default, they just generate text. But we would like to take that text. They're generating and how to do useful things like search the web internal databases.

34
00:07:22.040 --> 00:07:36.330
Kevin Martin: go up to Apis, you know. Maybe like make a dinner reservation for us. Book a haircut, whatever it is. We want them to be able to do to do things for us, and

35
00:07:36.920 --> 00:07:56.629
Kevin Martin: that's where Llm tools come in. So this was really originally pushed by Openai. So if you're if you've dealt with model tools. You probably are dealing with them from Openai, because they're just the most mature at this. But in here we can actually go.

36
00:07:56.780 --> 00:08:05.750
Kevin Martin: have chat, gpt up. We can do a new chat. And we can say, like, what is

37
00:08:06.070 --> 00:08:08.689
Kevin Martin: yeah, what is the

38
00:08:10.580 --> 00:08:17.910
Kevin Martin: current? Oh, my, gosh, I'm gonna expose some ignorance here, what is the current standings in the Mlb.

39
00:08:18.040 --> 00:08:32.290
Kevin Martin: And it can go and use a web search tool. It will go and pull it will go and pull current rankings in major league baseball and give me a response. Right? So this is this is basic tool use. But again, this is

40
00:08:32.780 --> 00:08:38.009
Kevin Martin: this is web search. This is one tool. We might want internal databases, Apis.

41
00:08:38.320 --> 00:08:41.429
Kevin Martin: our Crm tools, all all kinds of stuff.

42
00:08:42.645 --> 00:08:45.490
Kevin Martin: So yeah, this is

43
00:08:45.810 --> 00:08:53.610
Kevin Martin: this is where, okay? We want, we have like a tool with the web search. We want more tools. The problem is

44
00:08:54.570 --> 00:09:02.219
Kevin Martin: tool use is kind of 1st generation. Tool use, let's say, is fragmented. So

45
00:09:03.520 --> 00:09:12.210
Kevin Martin: open AI has a different way of defining tools for use by the large language model.

46
00:09:13.841 --> 00:09:24.460
Kevin Martin: Google does. Then anthropic Claude does. So, all of them, all of them just have slightly different specifications.

47
00:09:25.070 --> 00:09:32.869
Kevin Martin: And for end users, you just go. Okay? So what? The the. So what is

48
00:09:34.050 --> 00:09:42.800
Kevin Martin: developers are are only going to end up putting effort into developing tools for

49
00:09:43.760 --> 00:09:55.295
Kevin Martin: for you know, the big, the big, large language models. That means that, you know, as maybe open AI gets more and more market share, they'll go in and

50
00:09:55.980 --> 00:10:10.899
Kevin Martin: take advantage of that by charging, charging higher prices because they know. Oh, yeah, we've got this company buying, buying our saw, our access to our large language model. They've spent time to develop tools so that our large language model can

51
00:10:11.500 --> 00:10:33.989
Kevin Martin: interact with their Crm and get them important business data. Or you know, it can interact with their business intelligence dashboarding thing to develop dashboards. And okay, we've got them locked in. We can. We can keep continue charging higher prices. So it gives you as the user less flexibility. When the tools are

52
00:10:34.200 --> 00:10:43.020
Kevin Martin: kind of more bespoke, it also means that the tools are more prone to break so historically.

53
00:10:43.370 --> 00:10:50.190
Kevin Martin: historically as in a few months ago. Right? 1st generation tools they're

54
00:10:50.630 --> 00:10:54.649
Kevin Martin: their design and implementation is largely on

55
00:10:55.130 --> 00:11:03.150
Kevin Martin: the client side. So if I wanted to let my model. Use my Crm.

56
00:11:03.430 --> 00:11:06.589
Kevin Martin: Then I would have to like

57
00:11:06.710 --> 00:11:20.619
Kevin Martin: have have you know, in Api from my Crm, so I would have the salesforce Api. I would go on my end, write custom scripts

58
00:11:20.980 --> 00:11:22.400
Kevin Martin: so that

59
00:11:22.872 --> 00:11:44.089
Kevin Martin: so that my model can essentially call those scripts. I would write the little interface between the model and those scripts the model would call like a set of functions that I custom defined to go interact with the salesforce api and then I would get some kind of response, and it would work back and forth that way.

60
00:11:44.090 --> 00:11:54.020
Kevin Martin: But that means that if anyone changes anything in there the model breaks. So if the model provider decides, hey, we're going to change

61
00:11:54.530 --> 00:12:18.469
Kevin Martin: the way our tool calling works, which is a fast moving area. They've done it several times in the past few months. Or if salesforce goes, hey, we're going to change our Api, then you as the user or end up having broken tools, your Llm. Your your cloud or your openai can no longer talk with.

62
00:12:19.012 --> 00:12:26.630
Kevin Martin: With your tools. You can't use it in the same way to track like, hey? What was the last thing this customer bought, or or whatever you want?

63
00:12:28.010 --> 00:12:29.110
Kevin Martin: And

64
00:12:29.561 --> 00:12:36.738
Kevin Martin: you have to go and get the developers to come in and and constantly be maintaining it. So large technical debt.

65
00:12:38.780 --> 00:12:50.718
Kevin Martin: so that's we have our our Llms. They're useful on their own things, like our octopus in a tank unable to interact with the outside world. You could also think of like

66
00:12:51.270 --> 00:12:59.560
Kevin Martin: for real nineties, kids you may remember, pdas, they were these little tiny computers

67
00:12:59.670 --> 00:13:07.029
Kevin Martin: that you walked around with. They're useful on their own, but they have pretty severe limitations. They

68
00:13:07.940 --> 00:13:13.590
Kevin Martin: they can do things like their calculator. You can put your appointments in them you can.

69
00:13:14.182 --> 00:13:21.197
Kevin Martin: You know, maybe do a language to language, but they are not going to be.

70
00:13:21.940 --> 00:13:30.270
Kevin Martin: they're not going to be connected to the Internet. They're not going to be able to install custom software kind of that. That was our state. That's why we came up with. Tools.

71
00:13:30.670 --> 00:13:35.450
Kevin Martin: Tools are kind of complicated. They end up being brittle and vendor specific.

72
00:13:35.800 --> 00:13:41.499
Kevin Martin: So we could have easily ended up in a situation where we had

73
00:13:42.900 --> 00:13:52.259
Kevin Martin: very different, like vendor specific ways of interacting with the Llm. So open AI could have come up with its own thing

74
00:13:52.857 --> 00:14:15.300
Kevin Martin: Google could have come up with its own thing, we would end up with a more fragmented ecosystem kind of think. You know, I have an iphone app. My iphone app does not work on an android or my Mac app does not work on a windows machine or a Linux machine. We might have had this more fragmented ecosystem.

75
00:14:15.460 --> 00:14:18.040
Kevin Martin: But Mcp came along.

76
00:14:18.150 --> 00:14:22.610
Kevin Martin: So Mcp is developed by Claude

77
00:14:23.060 --> 00:14:27.040
Kevin Martin: Claude or the makers of Claude. Rather, it's developed by anthropic.

78
00:14:27.240 --> 00:14:37.920
Kevin Martin: which is a team of folks who split off from Openai and the whole idea behind model context protocol is that

79
00:14:38.150 --> 00:14:43.040
Kevin Martin: they decided essentially, we're going to call a truce. We're going to come up with

80
00:14:43.150 --> 00:14:48.949
Kevin Martin: one way that all of our different models are going to interact with these tools.

81
00:14:49.080 --> 00:14:54.250
Kevin Martin: So they came up with, this robust specification.

82
00:14:54.510 --> 00:14:59.478
Kevin Martin: the idea is, okay, we're gonna come up with one specification

83
00:15:00.060 --> 00:15:05.105
Kevin Martin: any model that wants to work with. Let's say, salesforce.

84
00:15:05.780 --> 00:15:22.250
Kevin Martin: we're going to. We're just going to propose that everyone do this one thing and kind of miraculously, it looks like this one way, this one tool, this one thing, this model context protocol is actually being adopted.

85
00:15:23.870 --> 00:15:29.753
Kevin Martin: So yeah. The idea is that salesforce or

86
00:15:32.900 --> 00:15:40.661
Kevin Martin: I'm trying to think of companies that actually have these. So Atlassian has an open model context, protocol

87
00:15:41.720 --> 00:15:50.477
Kevin Martin: not an Api, but kind of the model context, protocol equivalent of an Api. They have an endpoint on the Internet that the models can go talk to.

88
00:15:50.910 --> 00:15:58.670
Kevin Martin: Github has this. A few major vendors have already adopted this, and we have.

89
00:15:59.040 --> 00:16:00.139
Kevin Martin: They have.

90
00:16:00.610 --> 00:16:19.109
Kevin Martin: It looks like the major model providers are are coming around. So Openai has said they will support it. They're starting to roll out support for model context protocol. Google has said, they'll support it. And again, in in

91
00:16:19.560 --> 00:16:23.219
Kevin Martin: in baby steps, they're starting to implement support for it.

92
00:16:23.570 --> 00:16:31.239
Kevin Martin: And yeah, the idea is, instead of the developer. For your company

93
00:16:31.808 --> 00:16:37.900
Kevin Martin: going, hey? We want to connect our Crm, let me go and write a bunch of custom functions

94
00:16:38.350 --> 00:16:48.859
Kevin Martin: you can just go and take your company's like large language model, whoever your provider is, and you can point it at an endpoint, or

95
00:16:48.860 --> 00:17:13.430
Kevin Martin: you could develop kind of these reproducible local tools. And again, no matter what your provider is, you can go and download a script or something. You can go and point your model at that script, and you can get it to run. And again, that leaves you with the flexibility to determine. Hey, Gemini is really good at this type of task.

96
00:17:13.460 --> 00:17:18.930
Kevin Martin: We're going to use Gemini, for you know, our accounting team. So

97
00:17:19.000 --> 00:17:24.347
Kevin Martin: our accounting team can point at our model context protocol server and our

98
00:17:25.470 --> 00:17:42.789
Kevin Martin: you know, our software development team, maybe for software development, there's that new open AI tool codex. So our software dev team is going to all to use a different tool. And they can all point at the same Mcp server. And they can all exist in happy harmony.

99
00:17:44.560 --> 00:17:53.551
Kevin Martin: so yeah, it standardizes like tool access. It actually lets models kind of discover the tools. So I can point.

100
00:17:54.570 --> 00:18:09.829
Kevin Martin: I can just update that server and every model that is pointing to that server can go. Oh, I have access to a new tool. The user can use this new tool. Now, and yeah, it puts security and permissions front and center. So

101
00:18:11.545 --> 00:18:14.100
Kevin Martin: excuse me, kind of a general

102
00:18:14.420 --> 00:18:17.710
Kevin Martin: shortcoming with large language models is.

103
00:18:18.322 --> 00:18:28.450
Kevin Martin: historically, they're developed by researchers who are very interested in coming up with the best, fastest.

104
00:18:29.110 --> 00:18:32.879
Kevin Martin: most comprehensive answering model. And they're

105
00:18:33.560 --> 00:18:45.259
Kevin Martin: they're considering this more of a research project or an academic pursuit rather than a piece of software that people are going to use. So security is kind of.

106
00:18:45.931 --> 00:19:02.748
Kevin Martin: again historically. Taken a little bit of a back seat model context protocol has mechanisms where, like me, when I sign in, I can give my Oauth token that I use with, you know Google or

107
00:19:03.600 --> 00:19:07.620
Kevin Martin: or Microsoft, or any of these kind of major

108
00:19:07.940 --> 00:19:22.559
Kevin Martin: security sign secure sign in providers, and I can go pass that off to model context protocol on the client on my machine, and that can go and pass it along to the servers. And whoever else needs it, so that you know

109
00:19:22.910 --> 00:19:26.789
Kevin Martin: upstream. I don't get access to anything that

110
00:19:27.350 --> 00:19:45.269
Kevin Martin: that I shouldn't have access to as a person, because it. You know, the the end tools are actually seeing. Oh, yeah, this is, this is Chat Gpt. But it's asking on behalf of Kevin, so it shouldn't have access to any documents that Kevin doesn't have. And again, that hasn't always been the case.

111
00:19:46.500 --> 00:19:47.250
Kevin Martin: Okay?

112
00:19:47.620 --> 00:19:52.313
Kevin Martin: So how does it actually work?

113
00:19:53.500 --> 00:20:00.910
Kevin Martin: very high level. You set up a model context protocol server.

114
00:20:01.070 --> 00:20:16.770
Kevin Martin: This can be run locally. We're going to show it running locally. It can also be it can also be something that is exposed from a provider. So Atlassian has just a

115
00:20:16.890 --> 00:20:27.780
Kevin Martin: essentially a website, a web endpoint that you can. The model context protocol can interact with or models with model context, protocol enabled can go and interact with it.

116
00:20:28.260 --> 00:20:40.120
Kevin Martin: So there is a server that is sitting there. The server has a manifest. So it has a list of tools it has. Hey? This is the version I'm running

117
00:20:40.625 --> 00:20:45.250
Kevin Martin: all kinds of like useful information that the model would need to know.

118
00:20:45.790 --> 00:20:49.789
Kevin Martin: The model will go, and you'll point it to the server. It'll say, Hey.

119
00:20:50.990 --> 00:20:56.720
Kevin Martin: what are you? What can you do? The server will go. This is who I am. This is what I can do.

120
00:20:57.960 --> 00:21:15.399
Kevin Martin: And then them, and then the client will go cool. I do want to use you. Let's let's get started. So there's this one to one connection there, kind of like your connection. When you connect to a website, there's a 1 to one secure connection.

121
00:21:16.610 --> 00:21:19.769
Kevin Martin: and then from then on what?

122
00:21:19.970 --> 00:21:27.409
Kevin Martin: What your model is actually doing. And maybe this is, maybe this is worth pointing out, is your model is just

123
00:21:29.200 --> 00:21:31.570
Kevin Martin: knows that, for example.

124
00:21:31.880 --> 00:21:38.580
Kevin Martin: I am connected to a model context protocol server that can tell me the weather forecast.

125
00:21:39.226 --> 00:21:44.599
Kevin Martin: So what I need to do to get the weather forecast is, I need to send it

126
00:21:44.700 --> 00:21:54.299
Kevin Martin: a location where I want the weather like a latitude and a longitude, so it will go and send a Json formatted text

127
00:21:56.010 --> 00:21:59.230
Kevin Martin: for those of you who don't know

128
00:21:59.910 --> 00:22:06.580
Kevin Martin: Json is. You've probably seen Json. It's here's here's a good one.

129
00:22:07.050 --> 00:22:35.230
Kevin Martin: Json looks like this. So it's the kind of curly braces it's got a bunch of like lists and lookup dictionaries to find inside of it. So it knows it needs to send a Json tech on my, you know, Claude, or Chat Gpt or Gemini knows. Okay, I need to send like these 2 fields, latitude and longitude to the server. The server is going to do something with that, and it's going to send me back

130
00:22:35.440 --> 00:22:40.739
Kevin Martin: the weather, and then I can incorporate that weather into whatever the user has asked me.

131
00:22:42.660 --> 00:22:51.014
Kevin Martin: And yeah, the one of the nice things is this, the

132
00:22:51.920 --> 00:23:01.309
Kevin Martin: how how the request is is handled, and security checks, and anything you need is handled as a part of this kind of handshake

133
00:23:01.420 --> 00:23:07.329
Kevin Martin: between the client and the server. Between your machine, the client machine and the server, which is,

134
00:23:08.880 --> 00:23:18.597
Kevin Martin: some kind of remote service that you're calling. And you know, if I, for whatever reason, do not have access to

135
00:23:19.160 --> 00:23:20.740
Kevin Martin: to request

136
00:23:21.050 --> 00:23:30.079
Kevin Martin: the weather. Then the server will tell me. Oh, sorry you're denied, and then my model can tell me, hey? It looks like I don't have access for whatever reason.

137
00:23:32.070 --> 00:23:41.150
Kevin Martin: And yeah, everything is auditable standardized across vendors, platforms, etc.

138
00:23:41.360 --> 00:23:51.149
Kevin Martin: cool. So we. We talked about this a little bit, but the the exciting part of this is, it's bring your own tools for AI. So

139
00:23:53.570 --> 00:24:07.151
Kevin Martin: a large part of, for example, why App Apple maintains, like this crazy market dominance, and why those of us who use apple have this love hate relationship with it is

140
00:24:07.720 --> 00:24:23.250
Kevin Martin: once you're in, you're in. If I decided that oh, this android phone is actually a lot better. Well, you know, maybe a lot of the apps that I use regularly my Crm. In business terms

141
00:24:23.870 --> 00:24:29.420
Kevin Martin: will not port over to to android necessarily. So.

142
00:24:30.140 --> 00:24:38.440
Kevin Martin: Yeah, there is. There is a lot of there has historically been a lot of like lock in and software. And again, the

143
00:24:38.900 --> 00:24:47.899
Kevin Martin: the cool thing about model context protocol is there. They were very early on, and probably because no one has clear.

144
00:24:48.760 --> 00:24:52.770
Kevin Martin: no one has clear like overall market share.

145
00:24:53.680 --> 00:25:14.669
Kevin Martin: Very early on this kind of we'll call them like the 3rd place in the in the market share race group anthropic decided, hey, we're gonna do something so that we can all play together and then actually got adopted. So reduces vendor lock in reduces developer friction and maintenance. So I mentioned.

146
00:25:14.860 --> 00:25:16.429
Kevin Martin: like, yeah, if

147
00:25:16.960 --> 00:25:28.869
Kevin Martin: Github decides to change something upstream, then the custom function that your developer that your developer wrote so that your model can talk to Github will break.

148
00:25:29.380 --> 00:25:35.560
Kevin Martin: and you know Github might decide to push a chain, a breaking change every few months.

149
00:25:35.720 --> 00:25:39.070
Kevin Martin: and you know the way that your tool.

150
00:25:39.220 --> 00:25:51.590
Kevin Martin: your tool use works might change every couple of months, and there's no coordinating factor on there. There's not even the ability for your software as a service provider.

151
00:25:53.220 --> 00:26:02.129
Kevin Martin: your Github, to to actually update and maintain your your custom tools. So it there, there ends up being a lot of like technical debt.

152
00:26:02.680 --> 00:26:14.279
Kevin Martin: developed a lot of friction. And we can, we can just avoid that by using this model context protocol. So it pushes out integration type things to github.

153
00:26:14.610 --> 00:26:16.800
Kevin Martin: So I want it.

154
00:26:17.920 --> 00:26:25.669
Kevin Martin: Yes. So Github has a nice guide on how you connect to their Mcp server here

155
00:26:26.190 --> 00:26:36.819
Kevin Martin: or Atlassian also provides this. There are a couple of big providers and yeah permissions. Security is taken care of as part of the protocol.

156
00:26:38.690 --> 00:26:39.750
Kevin Martin: So

157
00:26:40.000 --> 00:26:46.100
Kevin Martin: that's the great things about it. And it does. It does actually look like it's being adopted. But

158
00:26:46.590 --> 00:27:11.589
Kevin Martin: just kind of your your reasons to be like, okay, we should. We should be very interested about this and continue watching it. But like it might not solve everything just yet. It's very, it's very young. So it was only announced in November of 2024. So only about half a year ago was this, did this come out? In the 1st place?

159
00:27:13.210 --> 00:27:22.001
Kevin Martin: they said, Google, open AI, and like, the other big ones have basically all said, oh, yeah, we'll support this.

160
00:27:22.740 --> 00:27:29.059
Kevin Martin: it's not super widely implemented yet, though. There are things like.

161
00:27:29.740 --> 00:27:36.520
Kevin Martin: yeah. Openai will allow you to connect to Mcp servers, but it will only allow you in their Api

162
00:27:37.210 --> 00:27:39.800
Kevin Martin: not yet in the chat. Gpt app

163
00:27:40.860 --> 00:27:51.794
Kevin Martin: it! I guess there might be a way to do it if you have an enterprise account I was seeing. But yeah, there's they're they're slowly rolling it out. It's not. It's not fully rolled out everywhere.

164
00:27:52.340 --> 00:27:55.750
Kevin Martin: Claude Claude is the

165
00:27:56.020 --> 00:28:08.666
Kevin Martin: language model that is developed by anthropic, which is the people who are pushing this. And you can't connect to remote remote servers. So if you wanted to connect to Atlassian or

166
00:28:09.720 --> 00:28:29.300
Kevin Martin: or Github right now and have your your cloud model, run, pull requests, and push, and and do all of the stuff that you can do with Github. You would need to have a paid subscription before you can connect that right now, so it's not available at like the lower tiers yet.

167
00:28:31.710 --> 00:28:38.379
Kevin Martin: Right now there's it's still up to you to determine is an endpoint safe. There's no kind of

168
00:28:40.200 --> 00:29:03.539
Kevin Martin: central certificate authority like there is with with the Internet in general, right? We have the Https certificate. That's a central central authority for helping determine like, is this good or safe, or whatever? Right now, there's not like a a core group who's responsible for like taking and maintaining the spec. And it's just fully in anthropic side project right now. So

169
00:29:03.810 --> 00:29:22.350
Kevin Martin: it could, it could potentially turn not not that it will. But there is some some potential. And yeah, but the adoption outlook is good. So Google, these are links to the various announcements where they said, Oh, yeah, we'll adopt model context protocol.

170
00:29:24.480 --> 00:29:25.687
Kevin Martin: Like, I said,

171
00:29:26.250 --> 00:29:33.820
Kevin Martin: anthropic Claude does support Mcp. You would think so because they're the ones who are developing it. But again.

172
00:29:34.580 --> 00:29:51.160
Kevin Martin: if you are running their desktop app, and you just want to connect on their desktop app and have kind of a demo with an established provider. They don't really. They don't really do that unless you're you have a a $20 a month or higher subscription tier.

173
00:29:53.750 --> 00:29:59.470
Kevin Martin: And then, yeah, what's needed for broader roll out just more

174
00:29:59.840 --> 00:30:05.609
Kevin Martin: more Mcp servers from software as a service. Let's see.

175
00:30:06.060 --> 00:30:18.430
Kevin Martin: they have. They have a list of folks who have adopted the protocol first, st or adopted the protocol. So far, so there's a bunch of them, as you can see. I think

176
00:30:20.720 --> 00:30:32.360
Kevin Martin: so. I kept on mentioning salesforce as like, oh, that's the that is the kind of

177
00:30:32.580 --> 00:30:46.277
Kevin Martin: customer management software that everyone use. Crm, there's that's what I'm thinking of. So salesforce is not on this list of people who have officially adopted it yet.

178
00:30:47.711 --> 00:30:54.480
Kevin Martin: tableau, another like huge one, still not quite on the list. So there is.

179
00:30:54.870 --> 00:30:59.001
Kevin Martin: But you know, Reddit or Redis rather so the big

180
00:31:00.420 --> 00:31:10.434
Kevin Martin: the big database solution from Amazon has adopted it. Lots of folks have there are still notable holdouts, though, let's say,

181
00:31:12.973 --> 00:31:31.939
Kevin Martin: yeah. You, definitely, we definitely want more adoption. But and yeah, better connections and existing tools like chat. Gbt, there right now, there's no way to adopt to add it in the chat. Gpt. Web app, even if you do have the paid subscription.

182
00:31:34.040 --> 00:31:40.049
Kevin Martin: So yeah, it. It's looking positive, but it's still, you know, tepid, positive. Let's say,

183
00:31:40.450 --> 00:31:47.020
Alex M.: There was a I just want to interrupt. Really, briefly, there was a question, the chat you could post, the link, what you were just sharing of all those companies.

184
00:31:47.020 --> 00:32:01.940
Kevin Martin: Yeah, thanks, Emil, I can totally post this link. So this is the official Github page for Mcp. So model context, protocol is an open source project right now, it's just

185
00:32:02.815 --> 00:32:10.000
Kevin Martin: being almost exclusively built open source by folks who work for anthropic.

186
00:32:10.760 --> 00:32:13.199
Kevin Martin: So right here,

187
00:32:25.140 --> 00:32:26.426
Kevin Martin: and then.

188
00:32:28.180 --> 00:32:54.310
Kevin Martin: yeah. So our, our takeaways from the the slides portion of this is Llm Llms need tools, but are kind of basic tool integrations where the developers are are actually writing custom code to connect to tools, leave leave to leave something to be desired. So we're

189
00:32:54.890 --> 00:33:01.340
Kevin Martin: you. You will just accumulate a lot of technical debt with these tools, and then you know, if

190
00:33:01.460 --> 00:33:11.249
Kevin Martin: if Joe is the one who developed your connector with so that your Llm. At your company can talk to

191
00:33:11.550 --> 00:33:31.150
Kevin Martin: salesforce is actually a great example here, because they don't have an Mcb server. So if if Joe wrote all of this code so that your Llm. Can talk to Salesforce and get the last quarter numbers and all of these important things that you need to know. And then, you know, Joe leaves, or you decide to go from

192
00:33:31.150 --> 00:33:48.909
Kevin Martin: Gemini to chat gpt. Then all of a sudden, all of that, all of that tools. All of that. All of those were the that work is kind of of dubious value to you, because Joe will need to update that. If Joe leaves the company.

193
00:33:49.720 --> 00:34:19.139
Kevin Martin: Then you know, someone else needs to take over that. Or you know, if you get a new model provider, then those 2, those same tools will not necessarily work, and Mcp goes and fixes that issue. It does so in a more secure way. Where, you know, there are actually folks on the project who are constantly thinking about security, whereas, you know, Joe, like me, might be more of a might be more of a data guy might be like, Oh, yeah, we'll make the connections and

194
00:34:19.230 --> 00:34:23.970
Kevin Martin: security might be a second second tier consideration. In that case

195
00:34:25.030 --> 00:34:32.950
Kevin Martin: adoption is early still, but the potential is big, and it does look like adoption is happening. It's just happening

196
00:34:33.340 --> 00:34:44.029
Kevin Martin: you it. It's actually happening pretty fast. So I think, the spec came out in November, Openai in March and Google in March. Both said that they would

197
00:34:44.883 --> 00:34:49.970
Kevin Martin: incorporate it, and they've been. They've been slowly incorporating it. But

198
00:34:50.451 --> 00:35:04.170
Kevin Martin: you know it. It's not fully there yet, I would say, give it based on. It's been 6 months, and you have halfway adoption at most of the big places. So given another few months, and it will be mostly adopted, I think.

199
00:35:06.149 --> 00:35:10.740
Kevin Martin: so definitely, definitely, a technology to keep your eyes on

200
00:35:11.610 --> 00:35:24.469
Kevin Martin: smoother smoother in many ways than designing your own tools. I have. We're gonna we're gonna share these slides at the end of at the end of this with you. There is a

201
00:35:24.880 --> 00:35:26.670
Kevin Martin: just a list of

202
00:35:27.293 --> 00:35:38.576
Kevin Martin: links that will be helpful to you as you go and do more research on this but in the intermediate time, in the last

203
00:35:39.360 --> 00:35:43.320
Kevin Martin: in the last part of class, I was gonna show.

204
00:35:44.830 --> 00:35:48.419
Kevin Martin: Excuse me, I was gonna show a demonstration of Mcp

205
00:35:50.830 --> 00:35:55.640
Kevin Martin: and how we can use it.

206
00:35:56.650 --> 00:36:04.410
Kevin Martin: So what I'm going to do, I'm gonna show it a couple of ways

207
00:36:07.536 --> 00:36:11.639
Kevin Martin: so let me get out of here

208
00:36:13.460 --> 00:36:16.809
Kevin Martin: and push this off to the side.

209
00:36:18.970 --> 00:36:26.210
Kevin Martin: And this is this is for the most part, there is these

210
00:36:26.370 --> 00:36:28.499
Kevin Martin: examples. So a quick start.

211
00:36:28.620 --> 00:36:35.299
Kevin Martin: server and client guide are going to be the majority of the code used in these examples. I

212
00:36:35.620 --> 00:36:43.210
Kevin Martin: I think I cut it down a little bit. But yeah, so the way that you can run this yourself is going to be

213
00:36:43.320 --> 00:36:47.240
Kevin Martin: spelled out inside of here.

214
00:36:47.880 --> 00:36:56.529
Kevin Martin: So the- that I'm actually going to skip

215
00:36:57.916 --> 00:37:04.499
Kevin Martin: this setup boilerplate stuff, cause I don't think that's particularly interesting here.

216
00:37:06.270 --> 00:37:10.600
Kevin Martin: So we just need to get down into

217
00:37:42.930 --> 00:37:43.750
Kevin Martin: hold on!

218
00:37:43.920 --> 00:37:46.099
Kevin Martin: It might be.

219
00:37:50.230 --> 00:37:55.250
Kevin Martin: Oh, no, yeah, it would be in file. Here, let me let me just double check right here.

220
00:37:55.360 --> 00:38:02.049
Kevin Martin: I think I have some of this set up. It might just be it might just be quicker to go and run through the.

221
00:38:03.680 --> 00:38:07.980
Kevin Martin: It might just go be quicker to run through the setup with you.

222
00:38:08.840 --> 00:38:15.609
Kevin Martin: Yeah, we'll just run through. We'll just run through this. It doesn't take that long, honestly, to get it set up. So maybe that's

223
00:38:16.575 --> 00:38:22.909
Kevin Martin: fine. So we're gonna we're gonna just run a couple a couple boiler plate.

224
00:38:23.530 --> 00:38:27.100
Kevin Martin: Oh, yeah, okay, sorry let me let me just

225
00:38:27.540 --> 00:38:33.269
Kevin Martin: let me, just before we get into like just doing this, demo ajit had a question

226
00:38:34.470 --> 00:38:45.049
Kevin Martin: ajit? Asked, is Mcpa an arm of agentic AI, the other being rag. Are there any other parts to Agentic AI. Is this the correct understanding? So.

227
00:38:46.570 --> 00:38:55.480
Kevin Martin: agentic? AI, I would say broadly, is providing your large language model

228
00:38:55.790 --> 00:39:00.640
Kevin Martin: with the ability to go and like

229
00:39:00.950 --> 00:39:04.990
Kevin Martin: it's providing it with some set of like tools

230
00:39:06.001 --> 00:39:18.069
Kevin Martin: and the ability to choose what tools are are useful in the moment, and then kind of like iteratively act on those tools. So I'd say this is

231
00:39:18.270 --> 00:39:23.419
Kevin Martin: adjacent to connected to a part of agentic, AI

232
00:39:24.038 --> 00:39:29.350
Kevin Martin: a gentech AI would be setting up the.

233
00:39:31.350 --> 00:39:33.340
Kevin Martin: It would be setting up the

234
00:39:34.270 --> 00:39:40.990
Kevin Martin: ability and the the logic for how your AI selects tools and whatnot, maybe.

235
00:39:41.250 --> 00:39:45.768
Kevin Martin: But it definitely needs tools to select from So

236
00:39:46.470 --> 00:39:51.029
Kevin Martin: means, let's say, this is part of agentic. AI rag is a closely related

237
00:39:51.440 --> 00:39:56.482
Kevin Martin: technology. You could, you could implement rag with tool calling. Rag is just

238
00:39:57.040 --> 00:40:08.270
Kevin Martin: going and giving your Llm. Access to external information before it goes, and release a a call back to the user. So

239
00:40:08.854 --> 00:40:12.650
Kevin Martin: a limitation of our large language models is.

240
00:40:12.920 --> 00:40:23.939
Kevin Martin: think the octopus in the jar right? So we went, and we taught on the octopus. Everything there was to know until November of 2024,

241
00:40:24.110 --> 00:40:29.549
Kevin Martin: and then at that point the octopus is sealed in the jar, and it doesn't have access to the outside world anymore.

242
00:40:29.700 --> 00:40:30.520
Kevin Martin: Right?

243
00:40:30.700 --> 00:40:39.109
Kevin Martin: It was. It was a baby. It learned. We trained it. And then we we put it away in this box. So same thing with large language models. It takes

244
00:40:39.640 --> 00:40:51.429
Kevin Martin: at this point to train, to train and fine tune a large language model like Chat Gpt. 4, 1 or 4, 5, or something like that. It takes

245
00:40:51.770 --> 00:40:54.069
Kevin Martin: a couple 1 million dollars.

246
00:40:54.640 --> 00:41:03.910
Kevin Martin: So like we, we spend all this time cramming information into this thing and then like it

247
00:41:04.320 --> 00:41:15.869
Kevin Martin: it. The training process takes a couple of 1 million dollars, so we can't just retrain it every day to give it access to new information. So what rag does is rag goes and

248
00:41:16.750 --> 00:41:37.430
Kevin Martin: basically perform search on a set of files and says, Okay, here is additional information that could be useful to you, based on what the user asked. And then the model will determine. Okay, is it worth incorporating this information into my response to the user? So if I asked a model. Oh, I can. I can show it with Claude, actually.

249
00:41:41.550 --> 00:41:45.410
Kevin Martin: So if I went to Claude

250
00:41:45.590 --> 00:41:47.818
Kevin Martin: and I turned off web search.

251
00:41:50.460 --> 00:42:00.070
Kevin Martin: ask it. This is very important. Web search is the tool it has access to. We don't necessarily want to have access to the tools. So if I ask it like, Hey.

252
00:42:01.058 --> 00:42:08.600
Kevin Martin: what was the top news story from last week?

253
00:42:10.889 --> 00:42:26.180
Kevin Martin: It won't know how to answer. It doesn't have access to the web, you're not. It was trained in January 2025. So everything past that is a is a black box to it. So that's what that's what tool use. And rag helps

254
00:42:26.990 --> 00:42:28.100
Kevin Martin: helps address.

255
00:42:30.320 --> 00:42:35.010
Kevin Martin: And I guess, yeah. Similarly, if I ask it, hey?

256
00:42:36.120 --> 00:42:42.919
Kevin Martin: what's the weather like in the Milchuck, Alaska?

257
00:42:44.670 --> 00:42:49.570
Kevin Martin: Beautiful Minil Chick, Alaska, where I am based at the moment

258
00:42:50.150 --> 00:42:55.357
Kevin Martin: I actually did give it access, I did give it access to a tool.

259
00:42:56.060 --> 00:43:03.980
Kevin Martin: So this is this is in an Mcp tool that I spun up.

260
00:43:06.350 --> 00:43:11.649
Kevin Martin: You can see actually nice under the hood. Exactly what happens. So

261
00:43:12.550 --> 00:43:21.850
Kevin Martin: I asked, what's the weather like in the No Chick, Alaska, it said, I can get you current weather for the no chick. Let me fetch that. The forecast for that location

262
00:43:22.640 --> 00:43:45.900
Kevin Martin: from its internal knowledge. It knows that the no chick is located at latitude 60 longitude 1 51 it goes and asks and sends that off to the Mcp server tool, the Mcp server tool gives it this response. This is just the the Noaa National Weather Service forecast.

263
00:43:48.640 --> 00:43:55.480
Kevin Martin: The model goes and parses that response, and it tells me what the weather is like today.

264
00:43:56.152 --> 00:44:05.420
Kevin Martin: Pretty typical for late June 54, with mostly cloudy chance of rain showers. That sounds like beautiful manil chick Alaska, that I know.

265
00:44:07.030 --> 00:44:21.880
Kevin Martin: the way. Okay, so let's go over the way that I set up this tool the way that I set up this tool. If you go to your Claude, you go to settings.

266
00:44:22.460 --> 00:44:25.030
Kevin Martin: you can actually click developer.

267
00:44:25.600 --> 00:44:31.939
Kevin Martin: And then, there is a config file here.

268
00:44:33.380 --> 00:44:38.699
Kevin Martin: It's actually open. Supply is easier on the eyes, in my opinion.

269
00:44:39.440 --> 00:44:47.630
Kevin Martin: So all I did was I pointed this to so long.

270
00:44:49.240 --> 00:44:55.020
Kevin Martin: I pointed this to a hold. On, there we go.

271
00:44:56.069 --> 00:44:56.469
Kevin Martin: Huh?

272
00:44:56.900 --> 00:45:02.320
Kevin Martin: I said, Okay, this is going to be the weather the weather tool.

273
00:45:04.310 --> 00:45:16.570
Kevin Martin: This is just some setup about what specific command is needed to run it on my machine. This is where it's located. So you can have. You can have your tools be

274
00:45:17.150 --> 00:45:34.570
Kevin Martin: local. So in this case, I actually did have to go and and spin up my own server the nice locally. Because I don't have the the paid version of of this, but you can. You can spin up your own server. It could be

275
00:45:35.574 --> 00:45:46.559
Kevin Martin: this could be just a central location on your company's network. And then, yeah, everyone at your company could go and point to this file and just run stuff from there.

276
00:45:47.700 --> 00:45:50.150
Kevin Martin: And in this case

277
00:45:50.977 --> 00:46:14.229
Kevin Martin: let me see, this is actually good, so that I can see where I put it the the supportings labs demo files. So it's pointing to it's pointing to this specific python file. And this python file is telling it exactly what it needs. And that's that's communicating with Noah, right? But this could easily be.

278
00:46:15.706 --> 00:46:24.869
Kevin Martin: This could easily be. In a general case, right? This could be pointing instead of to a local file. This could be pointing to

279
00:46:25.565 --> 00:46:50.640
Kevin Martin: like a an Mcp tool hosted by Noah itself, so that you know the burden of maintaining the tool is off of me. Even if it's being run locally, though a lot of these, a lot of these like types of connector tools are open source. So if you put the connector tool in a git folder.

280
00:46:50.900 --> 00:46:51.770
Kevin Martin: huh!

281
00:46:52.040 --> 00:46:58.390
Kevin Martin: That is connected to an open source project, all you would have to do is you would run a git pull

282
00:46:59.159 --> 00:47:15.220
Kevin Martin: kit is a a version management package. But you do get pull so just a 1 line command from within that folder, and you could go get the updated version of that tool which would be, you know, maintained by the the open source community.

283
00:47:15.640 --> 00:47:21.450
Kevin Martin: So that is an option, and then

284
00:47:23.760 --> 00:47:27.730
Kevin Martin: I will go and show alternatively.

285
00:47:28.637 --> 00:47:32.520
Kevin Martin: You can run this from the

286
00:47:33.780 --> 00:47:39.589
Kevin Martin: can run this from the command line. So I had, okay, I do want finder actually.

287
00:47:40.800 --> 00:47:44.065
Kevin Martin: So here.

288
00:47:48.560 --> 00:47:53.923
Kevin Martin: So it was Labs Quad, Mcp Demo. I didn't go deep enough

289
00:47:54.800 --> 00:47:59.893
Kevin Martin: So if I go in here right now

290
00:48:00.700 --> 00:48:08.980
Kevin Martin: right now I have this. In weather, this weather dot pi file, let's open it here.

291
00:48:09.400 --> 00:48:14.239
Kevin Martin: So this is this is just this is just a bunch of code that goes. And

292
00:48:14.715 --> 00:48:41.860
Kevin Martin: you can see it does. It? Actually does, you know, maybe maybe quite a bit of stuff. It goes and takes my existing my latitude and longitude. It passes it off to some kind of web service. It handles. It handles error. It does it. It does a bit of stuff, but that's ideally, if that if this is hosted by someone else, that's kind of abstracted away from you as the user, and you can just connect it.

293
00:48:43.490 --> 00:48:52.469
Kevin Martin: But so dumb, what we might want is

294
00:48:55.620 --> 00:49:01.250
Kevin Martin: go here, I think, yeah, client dot pi is here.

295
00:49:01.450 --> 00:49:04.320
Kevin Martin: So if I run.

296
00:49:04.910 --> 00:49:13.394
Kevin Martin: I basically, basically, I was just gonna show you you can run this from your command line as well pretty easily

297
00:49:14.780 --> 00:49:17.524
Kevin Martin: so this could be.

298
00:49:20.890 --> 00:49:31.635
Kevin Martin: yeah, Hi, and then it's just this big past path to our weather server.

299
00:49:34.820 --> 00:49:37.529
Kevin Martin: It doesn't have to be this. Hold on

300
00:49:39.990 --> 00:49:45.319
Kevin Martin: oh, man, I think I need to run it from inside of my

301
00:49:47.730 --> 00:49:51.499
Kevin Martin: might need to run it from inside of my container.

302
00:49:53.144 --> 00:50:08.770
Kevin Martin: Well, I mean you. You already stock it connected to the cloud app. All it's doing is the same thing. But from the client from the terminal, so you can. You can go run that on your own. Let's let's just open the floor to questions as we have them. And then Alex had a

303
00:50:08.840 --> 00:50:31.149
Kevin Martin: so Alex had a closing slide or 2. We'll go. We'll go and answer remaining questions and then you will have access to this slide deck I was using, and there is a like Lab Guide, which will show you like, okay, how do you actually run this this demo on your own machine?

304
00:50:32.671 --> 00:50:41.530
Kevin Martin: But let's see. Questions, how do we make? How do we scale local servers? If we have a very high

305
00:50:41.830 --> 00:50:49.380
Kevin Martin: load, it could be very resource, intensive. I don't think okay.

306
00:50:49.980 --> 00:51:08.029
Kevin Martin: It. It depends. It depends the most. The most resource. Intensive part of this would be would be hosting the model. Right? So all we're all we're doing here is in this

307
00:51:08.640 --> 00:51:31.339
Kevin Martin: python code that is running the whatever all we're doing in the python code that's running the server, though, is we're taking in a we're taking in a very small json. So just a small bit of text. And we're sending that out to, we're sending that out to a web address, and then we're getting the web. Address the response back

308
00:51:31.880 --> 00:51:40.490
Kevin Martin: so the model running the model is separate from from running the the Mcp. Server. If you were

309
00:51:40.990 --> 00:51:53.690
Kevin Martin: like an enterprise developer developing an Mcp endpoint, then you would. You would, of course, need, like, you know, maybe a load balancer in front of the requests that are coming in to help break it up.

310
00:51:53.980 --> 00:51:56.440
Kevin Martin: But you wouldn't.

311
00:51:56.570 --> 00:52:01.879
Kevin Martin: Yeah. The the actual Mcp services part can be quite small.

312
00:52:02.180 --> 00:52:04.070
Kevin Martin: So yeah it

313
00:52:05.090 --> 00:52:14.747
Kevin Martin: I. I don't imagine you would get into like for moderate deployments, for like lightweight tools like the weather server. I I don't imagine you'd get

314
00:52:16.400 --> 00:52:20.660
Kevin Martin: deep run into that. And then you're for more

315
00:52:21.410 --> 00:52:36.179
Kevin Martin: intensive things like if you wanted to connect to a SQL. Service or to a sequel service or something like that. It would just be your normal your normal constraints on people trying to connect to a SQL. Service.

316
00:52:36.570 --> 00:52:40.510
Kevin Martin: The the model is separate from the the tools they're using.

317
00:52:43.261 --> 00:52:47.530
Kevin Martin: Any other questions we have outstanding.

318
00:52:47.640 --> 00:52:50.750
Kevin Martin: Yeah, we'll we'll share the code for this afterwards.

319
00:52:56.860 --> 00:53:01.366
Kevin Martin: Okay, so high level points

320
00:53:02.900 --> 00:53:19.040
Kevin Martin: The models themselves are useful. Think like your pdas from the nineties, they can do stuff that that we think is useful by themselves, but we would like them to do more right. We would like to be to like instead of

321
00:53:19.709 --> 00:53:31.380
Kevin Martin: instead of like in general, asking them like, How do I book a campsite at the National Park? I want it to just book it for me, right? And so

322
00:53:31.420 --> 00:53:54.279
Kevin Martin: you would need tools for it to do that you'd need, like, maybe Internet search so that it can look for the right website. You'd need some kind of like browser integration so it can click the right buttons so you'd need tools for it to do that. We had a 1st generation of tools, but they are like more fragile because they don't.

323
00:53:54.580 --> 00:53:57.890
Kevin Martin: Yeah. The the 1st generation tools are just

324
00:53:58.970 --> 00:54:14.089
Kevin Martin: need to be. Custom bespoke creations by your own developers for your own software and model context, protocol goes and puts that development work on

325
00:54:14.912 --> 00:54:31.290
Kevin Martin: your your software as a service provider. So if you are connecting to Atlassian services. Atlassian is a great one, because they actually do have model context. Protocol servers up right now, you can just point your Llm at Atlassian's model context protocol server.

326
00:54:31.850 --> 00:54:48.319
Kevin Martin: You can provide it with like your credentials. And then anything that like you a jeep have access to your Llm. Will have access to on the server, so it can see. Oh, yeah, these are your projects that you have right now. A jeep

327
00:54:49.090 --> 00:55:06.010
Kevin Martin: it won't give you it won't give you projects that Emil only has access to it, will it will go and automatically kind of scope and limit that. And the onus for keeping that tool up to date gets shifted from

328
00:55:06.070 --> 00:55:28.710
Kevin Martin: your company and your it team to Atlassian. So Atlassian will go and maintain that, maintain that endpoint like they maintain like an Api endpoint that you might use otherwise. Alright, that is, I'm gonna say, that's that, Alex. I know you had a slide or 2 to close out, and then we'll we'll send you a follow up email

329
00:55:29.410 --> 00:55:31.740
Kevin Martin: when yeah, after class.

330
00:55:31.740 --> 00:55:54.139
Alex M.: Absolutely. Thank you, Kevin. I really appreciate it and enjoyed today's session. So I'm throwing some links in the chat for everyone. The 1st 2 I just shared are upcoming webinars relative to AI. But they're just, you know, different topics that we'll be covering one with sort of a gen AI toolkit. We'll be sharing a lot of great materials with that and resources. So I highly recommend registering for that one July 10, th

331
00:55:54.170 --> 00:56:10.429
Alex M.: and then the second one. We'll be following up with is AI in use. How generative AI is transforming today's systems. And secondly, I also have been referencing some courses that you can follow up with here. So one that's probably best suited for a follow up post. This

332
00:56:10.690 --> 00:56:34.769
Alex M.: presentation depending on where you are in terms of your skill level. It is building a gentic AI with model context protocol, that's GAI, 2, 1 0 4 you can find on our site. I also included that link in our chat below and make sure that Link works. Yep. And then we also have a QR code here that will just redirect you if you do happen to scan that to a contact us form

333
00:56:34.780 --> 00:57:02.109
Alex M.: we are available to do dedicated private courses as well as data dedicated webinars like I mentioned previously for your team, or even for your organization at scale. So that's something that we also offer, and we'd be happy to do so. So please reach out and we can follow up with you. And besides that, I will be sending a follow up email. Please keep an eye out for tomorrow. Recording materials from today's session will all be included in that. And again, a post

334
00:57:02.250 --> 00:57:13.990
Alex M.: Webinar survey. We'd love to hear your thoughts and see what you might be interested in moving forward. Any any like, you know any other topics that you're currently interested in. But you wanted to have a little bit more of a deeper scope on.

335
00:57:14.130 --> 00:57:29.609
Alex M.: we'd love to dive a bit deeper. So with that said, I think we'll we're just about at the end of the 2 Pm. Slated end time. So thank you all for joining today. And, Kevin, thank you for presenting it was it was a great session. So with that everyone. I hope you have all great day.

336
00:57:30.960 --> 00:57:32.019
Alex M.: Bye, Kevin.

The Data Must Flow: How Model Context Protocol (MCP) Accelerates AI Development [Webinar]

What You'll Learn:

Alert!

Modal Title

Error!

Default Title

Prompt

Confirm

Login